PyBench contains a lot of different tests and executes them not too long. Random fluctuations of a few percent are possible. We need a more stable realistic test working with floats.

$ ./python -m timeit 'x,y,d=1,0,0.01
for i in range(628):
  x,y=x-d*y,y+d*x'

Without patch: 1000 loops, best of 3: 220 usec per loop

With patch: 1000 loops, best of 3: 232 usec per loop

Difference is 5%.